CN105528792A - Medical image registration hybrid algorithm - Google Patents

Abstract

The invention provides a medical image registration hybrid algorithm, belongs to the technical field of medical image processing and specifically relates to a medical image registration method with a sparisity adaptive regularized matching pursuit method being combined. The algorithm provided in the invention combines the ideas of sparisity self-adaption and regularization, so that sparisity does not need to be taken as a priori condition in the image reconstruction process, and the problem that the sparisity must be known in the reconstruction process with the regularized matching pursuit method is solved. The sparisity estimation method is improved; compared with the similar method, the hybrid algorithm has better reconstruction effect on complex-texture images; and the hybrid algorithm has high practical applicability.

Description

A kind of medical figure registration hybrid algorithm

Technical field

The present invention is a kind of medical figure registration hybrid algorithm in conjunction with degree of rarefication adaptive canonical orthogonal matching pursuit method, belongs to technical field of medical image processing.

Background technology

The signal sampling rate that compressive sensing theory breaches conventional Nyquist theory calls is not less than the bottleneck of signal bandwidth 2 times, achieve signal sampling innovatively and compress and carry out simultaneously, decrease sampled data, save storage space, but include again enough quantity of information simultaneously.As long as just signal can be recovered accurately by method for reconstructing when needing.

CS theory mainly comprises the rarefaction representation of signal, linear measurement and method for reconstructing three aspects, and method for reconstructing, as the core of CS theory, obtains and pays close attention to widely.Method for reconstructing conventional is at present following several large class mainly: for l ₀the a series of greedy method that Norm minimum proposes, for l ₁the convex optimization method that Norm minimum proposes, iteration method and the minimum full variational method etc.Greediness method due to its have calculated amount little, rebuild effective and more easily realize, be most widely used.At match tracing method (Matchingpursuit, MP) on basis, orthogonal matching pursuit method (OrthogonalMatchingPursuit, OMP), the orthogonal matching process (RegularizedOrthogonalMatchingpursuit of canonical, ROMP), compression sampling match tracing method (CompressiveSamplingMP, CoSaMP), subspace method for tracing (SubspacePursuit, SP), degree of rarefication Adaptive matching method for tracing (SparisityAdaptiveMatchingPursuit, SAMP) is proposed successively.Canonical match tracing method needs degree of rarefication as prior imformation when rebuilding, but in practical application, degree of rarefication is normally unknown; Degree of rarefication Adaptive matching method for tracing can solve the situation of degree of rarefication the unknown, but its iteration step length arrange and unreasonable, be difficult to the convergence ensureing reconstruction signal process.

Summary of the invention

The object of the present invention is to provide the medical figure registration hybrid algorithm of a kind of set degree of rarefication adaptive canonical match tracing method that flow process is simple, Exact Reconstruction rate is high.

The object of the present invention is achieved like this:

Step 1: input parameter: perception matrix Φ, measuring-signal y;

Step 2: initialization: surplus r ₀=y, reconstruction signal x _rec=0, initial sparse degree K ₀=1, iterations n=0, indexed set atom set ${\mathrm{\Φ}}_{{\mathrm{\Γ}}^0}={\mathrm{\Φ}}^{T}y,$ Threshold epsilon;

Step 3: Γ ⁰=| g ₀| front K ₀individual maximal value index };

Step 4: if then K ₀=K ₀+ 1, go to step 1;

Step 5: use surplus r _ncalculate related coefficient u with each inner product arranged in perception matrix Φ, and find K from u ₀index value corresponding to individual maximal value is stored in J;

Step 6: regularization is carried out to the related coefficient of the corresponding atom of index value in J, and by regularization result stored in set J ₀in, the related coefficient of this set Atom must meet | u (i) | and≤2|u (j) | (i, j ∈ J);

Step 7: upgrade indexed set Γ ⁿ=Γ ^n-1∪ { k} and atom set

Step 8: utilize least square to try to achieve approximate solution

Step 9: upgrade surplus r _n=y-Φ x _n;

Step 10: if || r _n-r||≤ε ₂, then stop iteration, otherwise r=r _n, n=n+1, goes to step 5;

Step 11: output parameter: reconstruction signal x _rec, surplus r _n.

Beneficial effect

The present invention proposes a kind of degree of rarefication adaptive canonical match tracing method.The method does not need using degree of rarefication as priori conditions in signal reconstruction process, self-adaptation can approach degree of rarefication accurately build support set, and the Exact Reconstruction of settling signal and Exact Reconstruction rate, higher than existing congenic method, have higher practical application.

Accompanying drawing explanation

Fig. 1 is method flow diagram of the present invention.

Fig. 2 is that the method for the invention and MP method, OMP method, ROMP method, CoSaMP method and DP method are to Y-PSNR (PSNR) comparison diagram of 256 × 256lena image.

Fig. 3 is that the method for the invention and MP method, OMP method, ROMP method, CoSaMP method and DP method are to the reconstruction error comparison diagram of 256 × 256lena image.

Fig. 4 is original image (left side) and the reconstruction image (right side) of the method for the invention when sampling rate M/N=0.5.

Specific embodiments

Below in conjunction with drawings and Examples, the present invention is described in further detail.

The present invention proposes a kind of medical figure registration hybrid algorithm in conjunction with degree of rarefication self-adapting regular match tracing method for reconstructing.First, creating Candidate Set according to choosing with sampled signal correlativity the atom being greater than set threshold value, secondly, utilizing regularization thought to carry out postsearch screening to Candidate Set, being incorporated to support set by screening the atom obtained; Finally, approaching and upgrading surplus original signal is completed by the linear combination of the atomic building in support set.Detailed process is as follows:

Initialization: the original state value of each parameter in setting sparse signal process of reconstruction;

Step 1, step 2: definition measured value is y, obtained by calculation matrix and signal inner product, chooses gaussian random matrix herein as calculation matrix.Note reconstruction signal is x _rec, initial surplus r ₀=y, perception matrix Φ, obtained by the sparse base inner product of calculation matrix and signal, choose wavelet basis herein as sparse base.Perception matrix Φ is with parameter (K, δ _k) meet RIP character.Initial sparse degree K ₀=1, index value set support set is designated as Φ _Λ, iterations n=1, threshold value threshold epsilon ₂=10 ^-8.

Step 3, step 4: the estimation of degree of rarefication, initial sparse degree K ₀=1, if then increase K successively ₀until inequality is set up;

Step 5: calculate iteration surplus r _nwith perception matrix Φ each inner product arranged and related coefficient u={u _j| u _j=<r, Φ _j>} (j=1,2 ..., N), Φ _jfor the jth of perception matrix arranges, from u, choose K ₀index value corresponding to individual maximal value is stored in J;

Step 6: carry out regularization test to the related coefficient of the corresponding atom of index value in J, namely the related coefficient of this set Atom must meet | u _i|≤2|u _j| (i, j ∈ J), then by regularization result stored in set J ₀in;

Step 7: upgrade indexed set Γ ⁿ=Γ ^n-1∪ { k} and atom set

Step 8, step 9: adopt least square method carry out Signal approximation and upgrade surplus: r _n=y-Φ _Λx _n;

Step 10: if || r _n-r||≤ε ₂, then stop iteration, otherwise r=r _n, n=n+1, goes to step 5;

Step 11: output parameter: reconstruction signal x _rec, surplus r _n.

Beneficial effect of the present invention is: the medical image for texture complexity gives a kind of method for reconstructing, and the method can approach the true degree of rarefication of image fast, and iterations is less than traditional method, and reconstruction quality is better than existing method.

Fig. 2 and Fig. 3 is that the method for the invention and MP method, OMP method, ROMP method, COSAMP method and DP method are to Y-PSNR (PSNR) comparison diagram of 256 × 256lena image and reconstruction error comparison diagram respectively.What object adopted is 256 × 256 lena image, choose gaussian random matrix as calculation matrix, wavelet basis as sparse base, the emulation experiment of image reconstruction that utilized Matlab software to carry out under different sampling rate.

Claims (1)

1., in conjunction with a medical figure registration hybrid algorithm for degree of rarefication adaptive canonical match tracing method, it is characterized in that:

(1) input parameter: perception matrix Φ, measuring-signal y;

(2) initialization: surplus r ₀=y, reconstruction signal x _rec=0, initial sparse degree K ₀=1, iterations n=0, indexed set atom set threshold epsilon;

(3) Γ ⁰=| g ₀| front K ₀individual maximal value index };

(4) if then K ₀=K ₀+ 1, turn 4;

(5) surplus r is used _ncalculate related coefficient u with each inner product arranged in perception matrix Φ, and find K from u ₀index value corresponding to individual maximal value is stored in J;

(6) regularization is carried out to the related coefficient of the corresponding atom of index value in J, and by regularization result stored in set J ₀in, the related coefficient of this set Atom must meet | u (i) | and≤2|u (j) | (i, j ∈ J)

(7) indexed set Γ is upgraded ⁿ=Γ ^n-1∪ { k} and atom set

(8) least square is utilized to try to achieve approximate solution

(9) surplus r is upgraded _n=y-Φ x _n

(10) if || r _n-r||≤ε ₂, then stop iteration, otherwise r=r _n, n=n+1, turns 6

(11) output parameter: reconstruction signal x _rec, surplus r _n.